The world of design is changing more rapidly than ever before. I had considered writing a book of everything I know about product design, but I think the title would have to be “How we used to make software back before AI”. (Sad trombone noise)
One of my least favorite things about AI is the way LLMs communicate. It’s all verbose text! One design principle I have is “People don’t read. They skim”. Yet, chatGPT, Gemini, Claude, etc; they all communicate with these long text messages. There is a problem when this is the only way we experience the power of AI.
There are better ways to communicate. Brevity is a skill AI needs to develop. I believe that human designers also need to become better at information experience design.
Information architecture has been around a very long time, but mostly it has been focused on how to organize large sets of data and how to arrange a single page of information. Additionally, user interface is the bread the butter of product design, but is generally focused on buttons, menus, forms, and other actions invented at Xerox Parc. Lastly, there is data visualization, but that is typically for charts and graphs. I feel there is a missing domain of design focused on using this new paradigm.
Information experience design
I’m coining this term because there is not an alternative that hits the specific area I am trying to address. (Maybe someone can suggest something better).
IxD (to me) is how to arrange response data for rapid consumption. The consumer may be a person or potentially an agent of another AI. It is about how to convey lots of info without forcing the user to read a book report by a bot.
Remember the DBA? They used to be worth their weight in gold in Silicon Valley. They would optimize queries and data structures to make gathering information faster and more scalable. They mainly worked in SQL and ERD diagrams. I feel there is a similar job emerging, but works in a different way.
Let’s take a few use case examples:
- Many agents are working on a software project. A dashboard of charts is not the right interface to tell the human what is going on and what they need to do. Also, a giant text report is too long to look at.
- A person asks an agent to make flight arrangements. How to compare the flights and choose one requires more than just a bunch of bullet points.
- An agent asks another agent for status. The specific method the second agent responds could make a big difference in final quality and also efficient token usage.
In each of these cases, someone needs to optimize how information is conveyed. This is not an engineering task. It’s a design task. You clearly need to have a decent technical background but an engineer or product manager will not make the output consumable the same way as a designer.
There is a battle happening now between PM, Eng, and Design to decide how the system works. Tools like lovable fool PMs into thinking they are making a nice interface. Claude is making engineers think that they don’t need anyone else. This battle will likely end up with a ton of terrible interfaces and confused users and agents.
Designers are not the most aggressive of the three groups, so they have a risk of being steamrolled. However, I believe that designers are best suited to hold the steering wheel of AI to make systems work for humans.
Without spinning off into a different post, my main point here is that the specific kind of design focused on how to take large sets of data and make it easier to consume. We are about to be inundated with text and we need people thinking about how to make it better.
Whatya think?